View presentation slides for this project here
This project aims to develop a predictive model to forecast hotel booking cancellations using a Kaggle dataset of hotel reservations. The goal is to accurately predict cancellations, helping hotels optimize inventory, staff, and revenue strategies.
The project involves data cleaning, exploratory analysis, feature engineering, model selection, and evaluation using metrics like accuracy, precision, recall, and F1-score. By analyzing the factors contributing to cancellations, the project seeks to provide valuable insights for the hospitality industry, enabling better resource management and improved customer satisfaction.
The hospitality industry faces significant challenges with booking cancellations, which can lead to lost revenue and lower occupancy rates. With cancellation rates rising to 40%, there's a clear need for a predictive model to forecast these cancellations accurately. Such a model would allow hotels to proactively address potential cancellations, optimize staff and inventory management, and offer targeted promotions to retain bookings. This project aims to provide hotels with a tool to enhance revenue strategies, increase customer satisfaction, and reduce the financial impact of cancellations, giving them a competitive edge.
To build the predictive model, I used the Hotel Reservations Dataset from Kaggle, which includes data on customer bookings. Key features include the number of guests, meal plans, parking requirements, room types, lead time, arrival dates, and market segments. The dataset has 36,275 rows and 19 columns, with the target variable being booking_status (1 for canceled, 0 for not). I applied OneHotEncoder to categorical features and used StandardScaler for numerical features to prepare the data for modeling.
I began by creating a Logistic Regression model using scikit-learn's LogisticRegression class. The model was trained on the X_train_transformed and y_train data. This baseline model estimates the probability of booking cancellations (booking_status) based on input features, providing a foundation for comparison. The model's performance metrics are as follows:
- Accuracy of 80.4% - The percentage of correct predictions.
- Precision of 74.74% - The percentage of true positive predictions among all positive predictions.
- Recall of 63% - The percentage of true positive predictions among all actual positive predictions.
- F1 score of 68.4% - The harmonic average of precision and recall.
- ROC AUC score of 0.76 - Reflects the model's ability to distinguish between cancellations and non-cancellations.
These metrics provide a baseline to assess model performance and highlight areas for potential improvement.
Baseline Model:
Achieved:
- 85% Accuracy
- 86% Precision
- 65% Recall
- 74.2% F1 score
- 0.8 AUC
Achieved:
- 90.2% Accuracy
- 88.6% Precision
- 81.4% Recall
- 84.8% F1 score
- 0.88 AUC
| Logistic | Decision Tree | Random Forest | |
|---|---|---|---|
| Accuracy | 80.4% | 84.8% | 90.2% |
| Precision | 74.7% | 86.3% | 88.6% |
| Recall | 63% | 65% | 81.4% |
| F1 score | 68.4% | 74.2% | 84.8% |
| AUC | 0.76 | 0.8 | 0.88 |
- Final Model Selection: I selected the Random Forest Classifier as the final model. This model offers a good balance between complexity and performance, achieving a precision score of 89%, which is a significant improvement over the Logistic Regression(74.7%) and Decision Tree models(86.3%).
- Key Factor - Lead Time: The primary factor influencing cancellations is lead time, which measures the number of days between booking and arrival. Longer lead times correlate with higher cancellation rates, as plans can change over time. Hotel management should consider monitoring lead times closely and potentially adjusting cancellation policies or offering incentives for early confirmations to reduce cancellations.
For a thorough analysis of this dataset, checkout this notebook
This project is licenced under the MIT Licence